Localization of defects in solar cells using luminescence images and deep learning

Defect detection is a critical aspect of assuring the quality and reliability of silicon solar cells and modules. Luminescence imaging has been widely adopted as a fast method for analyzing photovoltaic devices and detecting faults. However, visual inspection of luminescence images is too slow for the expected manufacturing throughput. In this study, we propose a deep learning approach that identifies and localizes defects in electroluminescence images. Images are split into 16 tiles prior to training and treated as separate images for classification. The classified tiles provide both defect labels and their positions within the cell. We demonstrate the use of this novel approach to replace visual inspection of luminescence images in photovoltaic manufacturing lines to achieve fast and accurate defect detection.